A large number of current applications requires situative, contextualized information management, as to select relevant information quickly and well targeted from the ever-increasing amount of data.

In addition to well-known challenges like integration of data from many sources or the interpretation of sensor data interpretation, we need to deal with the selection and adaptation of data based on the context as well as the recommendation of topics and possible actions. In order to be valuable for users, these steps need to be compute with very low latency. Besides performance engineering, our focus of our work has been on learning user profiles in situations where not many interactions are known. A promising way to address this challenge is to analyze text content produced or observed by users.